A Deep Learning Approach For Skin Cancer Image Classification And Detection Using CNN Yolov5 Algorithm

Skin cancer is a rapidly growing complex disease that increasing challenge to international health organization. Traditional methods for detecting skin cancer challenges are accurately identifying the type and severity of the disease through image processing techniques. To address these limitations, a novel Classification Convolutional Neural Network (CNN) with YOLOv5 algorithm is applied for precise skin image prediction. The HAM10000 dataset is used as the input source, and a Sharpening Spatial Filter (SSF) is applied as a filtering process to improve the sharpness of the images during preprocessing. For skin cancer segmentation, the Mean Shift Clustering (MSC) technique is used to identify high thickness areas in the affected region. The Optimal Region Function Neural Network (ORFNN) is then used for feature selection, removing multi-scale and non-redundant texture features from segmented images, which improves the distinction between malignant and benign lesions. CNNs are utilized for feature extraction to differentiate cancerous from benign lesions, while YOLOv5 enables proper detection and localization of skin cancer areas, ensuring accurate classification. The method categorizes skin cancer into four diseases types: Melanoma (MEL), Necrobiosis Lipoidica, Actinic Keratoses (AKIEC), and Dermatofibroma (DF). The proposed model demonstrates improved achievement with an accuracy 98.05%, precision 96%, specificity 96%, PPV 96.55%, and NPV 96.45%.